What can one expect when using MC for serious backtesting? I assumed I would be able to perform intraday backtests based on years of tick data but I keep running into problems.
Simple example: I've imported 1 year of tick data for a futures symbol (which in itself takes about 20m - 30m to do). This is only a few million ticks (the full data set spans many years and many instruments). I then open up a chart, set the date range for the chart for the full year of data, apply a strategy (with the "bar magnifier" on) and ... kaboom... nothing. No immediate feedback that I can see. Everything is sluggish. Finally, after several minutes the chart window is updated showing some strategy trades. As I try to scroll the chart to the start of data ... kaboom... nothing... until after several minutes, the chart jump-scrolls a bit. When I open the performance reports... kaboom... nothing... until after many minutes. Same thing when I switch reports, always sluggish and waiting, waiting, waiting.
I'm not using a super computer but there are free processor cores and at least a few GB of unused RAM while the test is running in MC (64 bit).
I hope I am missing something here or I've just bought a piece of software that is of no use to me.
Cheers,
TB
Backtesting on tick data - what to expect?
- TrueBalance
- Posts: 3
- Joined: 01 Jan 2014
- Has thanked: 2 times
- TJ
- Posts: 7744
- Joined: 29 Aug 2006
- Location: Global Citizen
- Has thanked: 1033 times
- Been thanked: 2223 times
Re: Backtesting on tick data - what to expect?
What is your computer spec? CPU, MHz, RAM, SSD?What can one expect when using MC for serious backtesting? I assumed I would be able to perform intraday backtests based on years of tick data but I keep running into problems.
Simple example: I've imported 1 year of tick data for a futures symbol (which in itself takes about 20m - 30m to do). This is only a few million ticks (the full data set spans many years and many instruments). I then open up a chart, set the date range for the chart for the full year of data, apply a strategy (with the "bar magnifier" on) and ... kaboom... nothing. No immediate feedback that I can see. Everything is sluggish. Finally, after several minutes the chart window is updated showing some strategy trades. As I try to scroll the chart to the start of data ... kaboom... nothing... until after several minutes, the chart jump-scrolls a bit. When I open the performance reports... kaboom... nothing... until after many minutes. Same thing when I switch reports, always sluggish and waiting, waiting, waiting.
I'm not using a super computer but there are free processor cores and at least a few GB of unused RAM while the test is running in MC (64 bit).
I hope I am missing something here or I've just bought a piece of software that is of no use to me.
Cheers,
TB
What OS?
- TrueBalance
- Posts: 3
- Joined: 01 Jan 2014
- Has thanked: 2 times
Re: Backtesting on tick data - what to expect?
Core I7, 2.66GHz, 8MB RAM, SSD for OS disk (and MC installation and files as far as I can tell)What is your computer spec? CPU, MHz, RAM, SSD?
What OS?
Windows 7
I take if from your response that what I experience is not "normal" behaviour? Maybe I should be looking for a fix at my computer/OS setup instead.
- TJ
- Posts: 7744
- Joined: 29 Aug 2006
- Location: Global Citizen
- Has thanked: 1033 times
- Been thanked: 2223 times
Re: Backtesting on tick data - what to expect?
I don't assume anything...Core I7, 2.66GHz, 8MB RAM, SSD for OS disk (and MC installation and files as far as I can tell)What is your computer spec? CPU, MHz, RAM, SSD?
What OS?
Windows 7
I take if from your response that what I experience is not "normal" behaviour? Maybe I should be looking for a fix at my computer/OS setup instead.
I ask the question because I need to know where to start. I have seen people using 2GB ram and ask why his MultiCharts is so slow.
Is this a notebook computer?
8 GB is not a lot of memory these days; OS takes up a large part, also depending how much file index you have on RAM, you might only have half the memory left. 8 GB is usable, but marginal for large dataset work.
If your expectation is speed, a large SSD would help, a fast CPU clock is important, because backtest speed is not based on the number of cores, but clock speed.
If your computer uses shared video memory, that would slow down your scrolling performance dramatically.
- TJ
- Posts: 7744
- Joined: 29 Aug 2006
- Location: Global Citizen
- Has thanked: 1033 times
- Been thanked: 2223 times
Re: Backtesting on tick data - what to expect?
What is your chart resolution?What can one expect when using MC for serious backtesting? I assumed I would be able to perform intraday backtests based on years of tick data but I keep running into problems.
::
... I then open up a chart, set the date range for the chart for the full year of data, apply a strategy (with the "bar magnifier" on) and ... kaboom... nothing. No immediate feedback that I can see. Everything is sluggish...
What other apps do you run in this computer?
What is "several minutes"?...Finally, after several minutes the chart window is updated showing some strategy trades.
What is/are your strategy? How intensive are the calculations?
Please post a screen capture of your TaskManager Performance tab during this operation.
What is "many minutes"?... As I try to scroll the chart to the start of data ... kaboom... nothing... until after several minutes, the chart jump-scrolls a bit. When I open the performance reports... kaboom... nothing... until after many minutes. Same thing when I switch reports, always sluggish and waiting, waiting, waiting.
How many lines of trade are in the reports?
Please post a screen capture of your TaskManager Performance tab during this operation.
Re: Backtesting on tick data - what to expect?
You pc looks average, but is definitely up to the task.
There isn't a platform on this planet that will give you fast tick by tick back-tests. In this instance, it isn't the tools you're using, it's the project you're trying to accomplish. Institutional platforms that cost 1000x MC's $1500 are probably slower.
Be careful of the bar magnifier. I've always had issues with it being reliable and recently found something interesting (been too busy to work with MC on this but I posted about it on another thread). My suggestion for this is to use a data1 of tick data and a data2 to provide signals. This is reliable.. yet still slow.
My suggestion: Don't build strategies that need a tick by tick test... I do it and it drives me crazy. Much easier to come up with simpler ideas and build those.
There isn't a platform on this planet that will give you fast tick by tick back-tests. In this instance, it isn't the tools you're using, it's the project you're trying to accomplish. Institutional platforms that cost 1000x MC's $1500 are probably slower.
Be careful of the bar magnifier. I've always had issues with it being reliable and recently found something interesting (been too busy to work with MC on this but I posted about it on another thread). My suggestion for this is to use a data1 of tick data and a data2 to provide signals. This is reliable.. yet still slow.
My suggestion: Don't build strategies that need a tick by tick test... I do it and it drives me crazy. Much easier to come up with simpler ideas and build those.
- TrueBalance
- Posts: 3
- Joined: 01 Jan 2014
- Has thanked: 2 times
Re: Backtesting on tick data - what to expect?
Thank's for the input, guys.
Because of my expectations I had a lapse of cognitive dissonance after the purchase I guess. I'm experimenting a bit and backtesting with MC will be viable although not exactly as I had hoped.
Cheers,
TB
Because of my expectations I had a lapse of cognitive dissonance after the purchase I guess. I'm experimenting a bit and backtesting with MC will be viable although not exactly as I had hoped.
Cheers,
TB
- Dave Masalov
- Posts: 1712
- Joined: 16 Apr 2010
- Has thanked: 51 times
- Been thanked: 489 times
Re: Backtesting on tick data - what to expect?
TrueBalance,
We failed to reproduce the behavior that you describe in a similar environment. Calcualtion on tick data can be resource demanding, but you should not experience several minutes delays.
Please come to our Live Chat tomorrow from 6:30 AM to 4 PM EST to demonstrate the problem: http://messenger.providesupport.com/mes ... pport.html
We failed to reproduce the behavior that you describe in a similar environment. Calcualtion on tick data can be resource demanding, but you should not experience several minutes delays.
Please come to our Live Chat tomorrow from 6:30 AM to 4 PM EST to demonstrate the problem: http://messenger.providesupport.com/mes ... pport.html